Clinical Value Of MR Radiomics In Predicting Microsatellite Instability In Rectal Cancer

Purpose:MR radiomics is used to predict the microsatellite instability?MSI?of rectal cancer to better direct clinical neoadjuvant therapy.Materials and methods:63 colorectal cancer patients of our hospital were confirmed the expression status of mismatch repair protein and microsatellite states by immunohistochemistry.37 patients were microsatellite stability?23 male patients,14 female patients,the average age was 58.05,the range of age is 38-82?and 26 patients were microsatellite instability?13 male patients,13female patients,the average age was 61.03,the range of age was 30-83?.First,histogram matching and homogenizing were performed on all the MR imaging,then two experienced radiologists delineated the rectal tumor in T₂ and ADC images and automatically extracted 385radiomics features including first-order histograms?42?,morphological features?9?and texture features?334?.The radiomics features were statistically analyzed to identify the differences between the two groups,and the potential predictive features to differentiate the MSI group from the MSS group were subsequently tested using multivariable logistic regression analysis.Result:MSI showed no significant difference in clinical features and morphological characteristics with MSS.In T₂ imagines,MSI and MSS showed statistically significant difference in texture features includingGLCMEnergy_angle135_offset1,GLCMEnergy_AllDirection_offset4_SD,Inverse Difference Moment_AllDirection_offset7_SD,GLCMEnergy_angle90_offset7ShortRunLowGreyLevelEmphasis_AllDirection_offset7_SD,LongRunLowGreyLevelEmphasis_angle135_offset7?P<0.05?.Combined with the all above features to predict MSI of rectal cancer,the area under curve?AUC?of receiver operating characteristic curve?ROC?was 0.805,accuracy?ACC?was 0.746,specificity was 0.692,sensitivity was 0.784,and the threshold was 0.490,95%CI:0.694-0.915?DeLong?.In ADC images,MSI and MSS of rectal cancer showed statistically significant difference in histogram features and texture features including:Quantile0.025,Quantile0.25,GLCMEntropy_angle0_offset4,LongRunHighGreyLevelEmphasis_angle0_offset7,LongRunLowGreyLevelEmphasis_angle135_offset7,LowGreyLevelRunEmphasis_AllDirection_offset4_SD,ShortRunLowGreyLevelEmphasis_AllDirection_offset4_SD?P<0.05?,Combined with the all above features to predict MSI of rectal cancer,the AUC was 0.877,ACC 0.867 Specificity of 0.923,0.824 sensitivity threshold is 0.582,95%CI:0.777-0.977?DeLong?.Conclusion:MR T₂ and ADC imaging radiomics features can predict the microsatellite status of rectal cancer.